← back to test matrix

Test 0.4

Does a line's own wiggle predict whether it's wrong?

Yes, moderately — a 0.62 correlation between wiggle and real error, using nothing but the line's own shape. It also flags a specific failure: 1.2% of lines are dead-smooth but nowhere near a real stripe — the model grabbing a batten instead.
What it means for SailScan

This is a free quality signal we can compute with no human label and no ground truth, layered on top of the confidence score from test 0.3. It won't replace confidence, but it catches a different kind of mistake: a confident, smooth line that's simply tracing the wrong thing on the sail.

0.62
wiggle-vs-error correlation
15
batten-grabs caught
1.2%
of all lines
The detail
MeasureValue
Wiggle-vs-error correlation (Spearman)0.62
Detections checked1091
"Batten-grab" lines caught (smooth line, far from any real stripe)15 of 1286 (1.17%)

We fit a smooth curve to every line the model drew (no human label needed) and measured how far the real points wander from that fit — "wiggle." Lines that are perfectly smooth but land far from any real stripe are almost always the model latching onto a batten instead of a stripe.

Source: 332-photo holdout, frozen 2026-08-25. 1,286 detections, all with valid curve fits.

raw data (JSON) →